home / skills / zephyrwang6 / myskill / task-drill

task-drill skill

/task-drill

This skill helps you decompose tasks into steps, assign AI or human roles, and provide workflow prompts for four task types.

npx playbooks add skill zephyrwang6/myskill --skill task-drill

Review the files below or copy the command above to add this skill to your agents.

Files (2)
SKILL.md
4.8 KB
---
name: task-drill
description: AI 任务钻头 - 任务拆解助手,能够指导人类按任务类型进行任务拆解,分配哪些任务应该给AI做,哪些任务应该是人类自己来做。支持四种任务类型:直接问题解决、直接输出生成、协作问题解决、协作输出生成,并为每种类型提供详细的工作流程和提示词策略。当用户提出任何任务、需要帮助制定计划或要求拆解工作时使用此技能。
---

# AI 任务钻头

## 任务拆解流程

当用户发给你任务后,按照以下步骤执行:

### 1. 任务分类
首先将任务分类为以下四种类型之一:

- **直接问题解决(Direct Problem Solving)**:快速解决具体问题,一次性交互获取答案
- **直接输出生成(Direct Output Generation)**:生成较长内容(论文、报告等),一次性交互
- **协作问题解决(Collaborative Problem Solving)**:通过多轮对话共同解决复杂问题
- **协作输出生成(Collaborative Output Generation)**:通过多轮对话共同生成或完善较长内容

### 2. 任务分析与分配
根据任务类型和特点,确定任务的执行主体(人类主导或AI主导)。

### 3. 详细拆解
将任务拆解为具体步骤,为每一步说明:
- 执行者(AI/人类)
- 具体操作说明
- 提示词建议
- 人类需要完成的事项

## 任务类型判断标准

### 直接问题解决
**特征**:有明确答案,信息相对完整,可通过一次性交互解决
**示例**:
- "如何优化这段代码的性能?"
- "这个API的参数说明是什么?"
- "帮我找出这个bug的原因"

### 直接输出生成
**特征**:需要生成较长内容,输入信息充足,可一次性完成
**示例**:
- "根据这份大纲写一篇2000字的技术文章"
- "基于这些数据生成一份分析报告"
- "根据要求写一个Python脚本"

### 协作问题解决
**特征**:问题复杂,需要多轮对话,信息不完整或需要逐步探索
**示例**:
- "帮我设计一个完整的系统架构"
- "我的机器学习模型表现不佳,帮我优化"
- "想创业但不知道做什么方向"

### 协作输出生成
**特征**:需要生成较长内容但信息不足,或需要逐步完善
**示例**:
- "帮我写一份商业计划书"
- "设计一个完整的用户手册"
- "创作一部小说的大纲和正文"

## 拆解输出格式

按照以下格式输出任务拆解结果:

```
任务名称:[任务名称]
任务类型:[直接问题解决/直接输出生成/协作问题解决/协作输出生成]
任务分配:[人类主导/AI主导]
为什么属于此类型:[解释为何任务属于该类型]
为什么要交给[人类/AI]:[解释分配理由]

### 任务拆解步骤:

**步骤1:[步骤名称]**
- 执行者:[AI/人类]
- 具体操作:[详细操作说明]
- 人类输入给AI的提示词:[根据任务类型优化的提示词]
- 人类需要做的事:[详细说明]

[继续列出后续步骤...]
```

## 提示词策略模板

### 直接问题解决类提示词
```
请直接回答以下问题:[问题描述]
要求:
1. 提供准确、简洁的答案
2. 如有多个解决方案,请比较优劣
3. 给出可操作的建议
```

### 直接输出生成类提示词
```
请根据以下要求生成[内容类型]:

背景信息:[背景]
具体要求:[详细要求]
输出格式:[格式要求]
篇幅要求:[字数/长度]
风格要求:[写作风格]

请生成完整的[内容类型]。
```

### 协作问题解决类提示词
```
我需要解决以下复杂问题:[问题描述]

目前已了解的信息:[已知信息]
面临的挑战:[具体困难]

请帮我:
1. 分析问题的核心要素
2. 提出可能的解决方案方向
3. 识别需要补充的关键信息
4. 制定下一步行动计划
```

### 协作输出生成类提示词
```
我将与你协作创建[内容类型],这是一个需要多轮完善的过程。

初步想法:[初步概念]
核心要求:[基本要求]
目标受众:[使用对象]

让我们先从确定整体框架开始,然后逐步完善细节。
请告诉我需要补充哪些信息才能更好地开始。
```

## 执行指南

1. **识别触发**:当用户提出任务、需要帮助或要求拆解工作时使用本技能
2. **快速分类**:根据任务特征快速判断属于哪种类型
3. **清晰分配**:明确每一步由AI执行还是人类执行
4. **提供工具**:为每一步提供具体的提示词模板
5. **保持灵活**:根据具体情况调整拆解方式和提示词

## 注意事项

- 确保任务拆解逻辑清晰,步骤之间有明确的依赖关系
- 提示词要具体、可操作,避免过于抽象
- 对于复杂任务,建议先制定整体框架再细化具体步骤
- 优先考虑人类的优势(创造力、判断力)和AI的优势(信息处理、模式识别)

Overview

This skill guides users through decomposing tasks and deciding which parts humans should handle and which parts AI should handle. It supports four task types: Direct Problem Solving, Direct Output Generation, Collaborative Problem Solving, and Collaborative Output Generation. The skill produces step-by-step breakdowns, role assignment, and ready-to-use prompt templates. Use it whenever you need a structured plan to execute or delegate work between humans and AI.

How this skill works

When given a task, the skill first classifies it into one of the four types based on scope, information completeness, and interaction needs. It then assigns each subtask to either AI or a human, explains the rationale, and produces a clear sequence of steps. For every step it supplies the executor, explicit actions, human-to-AI prompt suggestions, and any human deliverables required. Prompts follow optimized templates for each task type to make handoffs actionable.

When to use it

  • You need to decide which parts of a project an AI should perform and which require human judgment.
  • You want a reproducible, stepwise plan for solving a problem or producing long-form output.
  • A task seems complex or ambiguous and would benefit from multi-turn collaboration with AI.
  • You need concrete prompt templates to hand AI clear instructions for each subtask.
  • You’re preparing a workflow where responsibilities and dependencies must be explicit.

Best practices

  • Classify the task first; that determines interaction style and granularity of steps.
  • Prefer AI for data processing, pattern recognition, and first drafts; reserve humans for creativity, ethics, and final validation.
  • Start collaborative work with a high-level framework, then iterate in defined rounds.
  • Provide concise, context-rich inputs when calling AI to reduce clarification loops.
  • Document dependencies between steps so later stages can rely on earlier outputs.

Example use cases

  • Optimize a block of code: classify as Direct Problem Solving and provide a single-shot diagnosis plus suggested fixes.
  • Write a technical report from data: use Direct Output Generation with a detailed brief and format requirements.
  • Design system architecture: treat as Collaborative Problem Solving with iterative exploration and decision checkpoints.
  • Create a product manual: use Collaborative Output Generation to establish structure, then refine sections across rounds.
  • Plan a marketing campaign: decompose into research (AI), creative concepts (human+AI), and validation (human).

FAQ

How do I pick between direct and collaborative types?

If the task has a clear answer and enough input, choose Direct. If it needs back-and-forth, exploration, or progressive refinement, choose Collaborative.

Can steps switch executors mid-task?

Yes. A step can be AI-led for a draft and then handed to a human for review and finalization; document that handoff explicitly.

What makes a good prompt for each type?

Direct prompts should be specific and outcome-focused; collaborative prompts should include known constraints, open questions, and a request for next steps.